Sign-constrained synapses and biased patterns in neural networks
Raju Viswanathan · Journal of Physics A Mathematical and General · 1993
The storage of biased patterns is examined in neural networks with sign-constrained synapses. Every neuron has outgoing synapses which are either all inhibitory or all excitatory. For random patterns stored in such networks, it is known that the presence of a discrete gauge symmetry makes the maximal storage capacity independent of the proportion of excitatory neurons to inhibitory neurons. When the stored patterns are biased, however, this discrete gauge symmetry is broken, with the result that the maximal capacity depends on the proportion of excitatory neurons to inhibitory ones. The dependence of the capacity on the fraction of excitatory neurons in the network, f, is calculated using the space of interactions approach. It is found that the storage capacity is maximal at f=0.5; this result is true regardless of the particular value of the bias in the stored patterns. The significance of this result in the neurophysiological context is discussed.